AIC, BIC in MIXED

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

AIC, BIC in MIXED

Kornbrot, Diana
Hi

As I understand it, if ALL numbers in target variable are multiplied by a constant, goodness of fit should not change 
BUT
running MIXED with NPOS as target, normal, identity
gives 

Akaike Corrected
1926.495
Bayesian
1965.075
*Generalized Linear Mixed Models.
GENLINMIXED
  /DATA_STRUCTURE SUBJECTS=Participant REPEATED_MEASURES=Repeated1*Repeated2 COVARIANCE_TYPE=UNSTRUCTURED
  /FIELDS TARGET=FreqPos TRIALS=NONE OFFSET=NONE
  /TARGET_OPTIONS DISTRIBUTION=NORMAL LINK=IDENTITY
  /FIXED  EFFECTS=Repeated1 Repeated2 Between Repeated1*Repeated2 Repeated1*Between Repeated2*Between Repeated1*Repeated2*Between USE_INTERCEPT=TRUE
  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100 CONFIDENCE_LEVEL=95 DF_METHOD=SATTERTHWAITE COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0 SINGULAR=0.000000000001
  /EMMEANS TABLES=Repeated1 CONTRAST=NONE
   /EMMEANS TABLES=Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated2*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2*Between CONTRAST=NONE
  /EMMEANS_OPTIONS SCALE=TRANSFORMED PADJUST=LSD.
while running mixed on proportion = FreqPos/18 gives

Akaike Corrected
-223.942
Bayesian
-185.362
*Generalized Linear Mixed Models.
GENLINMIXED
  /DATA_STRUCTURE SUBJECTS=Participant REPEATED_MEASURES=Repeated1*Repeated2 COVARIANCE_TYPE=UNSTRUCTURED
  /FIELDS TARGET=Proportion TRIALS=NONE OFFSET=NONE
  /TARGET_OPTIONS DISTRIBUTION=NORMAL LINK=IDENTITY
  /FIXED  EFFECTS=Repeated1 Repeated2 Between Repeated1*Repeated2 Repeated1*Between Repeated2*Between Repeated1*Repeated2*Between USE_INTERCEPT=TRUE
  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100 CONFIDENCE_LEVEL=95 DF_METHOD=SATTERTHWAITE COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0 SINGULAR=0.000000000001
  /EMMEANS TABLES=Repeated1 CONTRAST=NONE
   /EMMEANS TABLES=Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated2*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2*Between CONTRAST=NONE
  /EMMEANS_OPTIONS SCALE=TRANSFORMED PADJUST=LSD.

HELP
Why are aic, bic not same for both analyses?
Any help greatly appreciated
best
Diana

_____________________________________
Professor Diana Kornbrot
Mobile
+44 (0) 7403 18 16 12
Work
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
+44 (0) 170 728 4626
[hidden email]
http://dianakornbrot.wordpress.com/
http://go.herts.ac.uk/Diana_Kornbrot
skype:  kornbrotme
Home
19 Elmhurst Avenue
London N2 0LT, UK
+44 (0) 208 444 2081                                                   
 ------------------------------------------------------------                                    





===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Reply | Threaded
Open this post in threaded view
|

Re: AIC, BIC in MIXED

Ryan
Diana,

Are any of the values on the original DV equal to zero (“0”)? 

Ryan 

On Feb 18, 2018, at 7:06 AM, Kornbrot, Diana <[hidden email]> wrote:

Hi

As I understand it, if ALL numbers in target variable are multiplied by a constant, goodness of fit should not change 
BUT
running MIXED with NPOS as target, normal, identity
gives 

Akaike Corrected
1926.495
Bayesian
1965.075
*Generalized Linear Mixed Models.
GENLINMIXED
  /DATA_STRUCTURE SUBJECTS=Participant REPEATED_MEASURES=Repeated1*Repeated2 COVARIANCE_TYPE=UNSTRUCTURED
  /FIELDS TARGET=FreqPos TRIALS=NONE OFFSET=NONE
  /TARGET_OPTIONS DISTRIBUTION=NORMAL LINK=IDENTITY
  /FIXED  EFFECTS=Repeated1 Repeated2 Between Repeated1*Repeated2 Repeated1*Between Repeated2*Between Repeated1*Repeated2*Between USE_INTERCEPT=TRUE
  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100 CONFIDENCE_LEVEL=95 DF_METHOD=SATTERTHWAITE COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0 SINGULAR=0.000000000001
  /EMMEANS TABLES=Repeated1 CONTRAST=NONE
   /EMMEANS TABLES=Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated2*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2*Between CONTRAST=NONE
  /EMMEANS_OPTIONS SCALE=TRANSFORMED PADJUST=LSD.
while running mixed on proportion = FreqPos/18 gives

Akaike Corrected
-223.942
Bayesian
-185.362
*Generalized Linear Mixed Models.
GENLINMIXED
  /DATA_STRUCTURE SUBJECTS=Participant REPEATED_MEASURES=Repeated1*Repeated2 COVARIANCE_TYPE=UNSTRUCTURED
  /FIELDS TARGET=Proportion TRIALS=NONE OFFSET=NONE
  /TARGET_OPTIONS DISTRIBUTION=NORMAL LINK=IDENTITY
  /FIXED  EFFECTS=Repeated1 Repeated2 Between Repeated1*Repeated2 Repeated1*Between Repeated2*Between Repeated1*Repeated2*Between USE_INTERCEPT=TRUE
  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100 CONFIDENCE_LEVEL=95 DF_METHOD=SATTERTHWAITE COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0 SINGULAR=0.000000000001
  /EMMEANS TABLES=Repeated1 CONTRAST=NONE
   /EMMEANS TABLES=Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated2*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2*Between CONTRAST=NONE
  /EMMEANS_OPTIONS SCALE=TRANSFORMED PADJUST=LSD.

HELP
Why are aic, bic not same for both analyses?
Any help greatly appreciated
best
Diana

_____________________________________
Professor Diana Kornbrot
Mobile
+44 (0) 7403 18 16 12
Work
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
<a href="tel:+44%201707%20284626" value="+441707284626" target="_blank">+44 (0) 170 728 4626
[hidden email]
http://dianakornbrot.wordpress.com/
http://go.herts.ac.uk/Diana_Kornbrot
skype:  kornbrotme
Home
19 Elmhurst Avenue
London N2 0LT, UK
<a href="tel:+44%2020%208444%202081" value="+442084442081" target="_blank">+44 (0) 208 444 2081                                                   
 ------------------------------------------------------------                                    





===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Reply | Threaded
Open this post in threaded view
|

Re: AIC, BIC in MIXED

Rich Ulrich
In reply to this post by Kornbrot, Diana

For least square models, the AIC and BIC can be computed from RSS, the Residual Sum of squares, ignoring a constant.  From the Wiki article on AIC -

   Because only differences in AIC are meaningful, the constant  C2 can be ignored,
   which allows us to conveniently take AIC = 2k + nln(RSS) for model comparisons.
   Note that if all the models have the same k, then selecting the model with minimum
   AIC is equivalent to selecting the model with minimum RSS—which is the usual
   objective of model selection based on least squares.

Your RSS is changed when you multiply the scale of the criterion. That does not matter when it comes to /comparing/  models.  A single AIC or BIC after dropping the constant has no useful interpretation at all.


-- 

Rich Ulrich


From: SPSSX(r) Discussion <[hidden email]> on behalf of Kornbrot, Diana <[hidden email]>
Sent: Sunday, February 18, 2018 7:06:59 AM
To: [hidden email]
Subject: AIC, BIC in MIXED
 
Hi

As I understand it, if ALL numbers in target variable are multiplied by a constant, goodness of fit should not change 
BUT
running MIXED with NPOS as target, normal, identity
gives 

Akaike Corrected
1926.495
Bayesian
1965.075
*Generalized Linear Mixed Models.
GENLINMIXED
  /DATA_STRUCTURE SUBJECTS=Participant REPEATED_MEASURES=Repeated1*Repeated2 COVARIANCE_TYPE=UNSTRUCTURED
  /FIELDS TARGET=FreqPos TRIALS=NONE OFFSET=NONE
  /TARGET_OPTIONS DISTRIBUTION=NORMAL LINK=IDENTITY
  /FIXED  EFFECTS=Repeated1 Repeated2 Between Repeated1*Repeated2 Repeated1*Between Repeated2*Between Repeated1*Repeated2*Between USE_INTERCEPT=TRUE
  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100 CONFIDENCE_LEVEL=95 DF_METHOD=SATTERTHWAITE COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0 SINGULAR=0.000000000001
  /EMMEANS TABLES=Repeated1 CONTRAST=NONE
   /EMMEANS TABLES=Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated2*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2*Between CONTRAST=NONE
  /EMMEANS_OPTIONS SCALE=TRANSFORMED PADJUST=LSD.
while running mixed on proportion = FreqPos/18 gives

Akaike Corrected
-223.942
Bayesian
-185.362
*Generalized Linear Mixed Models.
GENLINMIXED
  /DATA_STRUCTURE SUBJECTS=Participant REPEATED_MEASURES=Repeated1*Repeated2 COVARIANCE_TYPE=UNSTRUCTURED
  /FIELDS TARGET=Proportion TRIALS=NONE OFFSET=NONE
  /TARGET_OPTIONS DISTRIBUTION=NORMAL LINK=IDENTITY
  /FIXED  EFFECTS=Repeated1 Repeated2 Between Repeated1*Repeated2 Repeated1*Between Repeated2*Between Repeated1*Repeated2*Between USE_INTERCEPT=TRUE
  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100 CONFIDENCE_LEVEL=95 DF_METHOD=SATTERTHWAITE COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0 SINGULAR=0.000000000001
  /EMMEANS TABLES=Repeated1 CONTRAST=NONE
   /EMMEANS TABLES=Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2 CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated2*Between CONTRAST=NONE
   /EMMEANS TABLES=Repeated1*Repeated2*Between CONTRAST=NONE
  /EMMEANS_OPTIONS SCALE=TRANSFORMED PADJUST=LSD.

HELP
Why are aic, bic not same for both analyses?
Any help greatly appreciated
best
Diana

_____________________________________
Professor Diana Kornbrot
Mobile
+44 (0) 7403 18 16 12
Work
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
+44 (0) 170 728 4626
[hidden email]
http://dianakornbrot.wordpress.com/
http://go.herts.ac.uk/Diana_Kornbrot
skype:  kornbrotme
Home
19 Elmhurst Avenue
London N2 0LT, UK
+44 (0) 208 444 2081                                                   
 ------------------------------------------------------------                                    





===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD